Xeon Vs. Opteron Comparison

ERDTdiver

Honorable
Aug 7, 2012
26
0
10,530
Should've been more specific. I have no specific processors in mind, just the basic differences between the two, what is one better at than the other. I'll be using them for a workstation/server set-up. I'd like to do some heavy web programming while also using the computer as a server.
 

ERDTdiver

Honorable
Aug 7, 2012
26
0
10,530
I'd like a xeon or opteron due to all of the coding and compiling, theres going to be a lot of computing. Plus with a 24/7 server for a website or two and also many files, a server cpu i think would be necessary. For coding and computing power it looks like AMD is the way to go.
 

wireflight

Honorable
Mar 15, 2012
5
0
10,510
Check here for a succinct presentation of relatively current relevant info in RE your request: http://www.elnexus.com/articles/Opteron-Xeon-Benchmarks-2012-01.aspx

For me, a 2P (or better, 4P) Opteron solution makes the most sense; for a mobo, Asus is probably the low-price champ in the 2P arena; Supermicro (aka: A+) 2P and 4P boards are about twice what you'd expect to pay for a decent consumer board socketed for a single Sandy Bridge cpu.

Most boards for the 6200-series Opterons can use common, relatively inexpensive, unbuffered consumer-grade RAM; if you're worried about baking your memory in a 24/7/365 solution, Koolance offers a liquid-cooling kit for such RAM (and also has water blocks for the G34 and C32 Opterons).

The motherboards for the Opteron are almost entirely geared towards server applications; however, some Opteron workstation solutions are on the market using a 2P Supermicro mobo.

As to why an Opteron workstation might use a mobo made and marketed for servers, at this level, I don't think it's reasonable to simply dismiss such a choice as a concession to expedience:

My guess is that performance for at least most workstation applications using the AMD CPU is sufficiently great, and that optimization of the board for workstation performance would yield such small improvement that such optimization would not be a worthwhile deployment of the mobo-manufacturer's resources -- a fact somewhat aggravated both by consumer prejudice and by AMD's election to market the Opterons as "server" CPUs.

If cost is an issue, consider the socket C32 Opterons: Asus and Tyan have 2P motherboards for these CPUs; I can't recall right now, but there is at least one ATX mobo for dual socket-C32 Opterons; a 2P server/workstation build using 4200-series Opterons would be cost-competitive with a 1P consumer-grade Core i7-2600K machine and would offer performance much improved over the Intel product.

Some people would say that isn't a fair contrast: the i7 mobo is usually available with particular cutting-edge tech not present on a server mobo (favors the i7 if you need an interface not supported on the server board) or the 4200-series Opteron is a low-end server CPU being compared to a mid-range consumer CPU (favors the Opteron, especially where cost is important and "mainstream tech" provides adequate connectivity and real-world performance).

I don't perceive any disadvantage in RE complexity of construction or operation as differentiating the 2P Opteron solution from any competently constructed and operated 1P machine (regardless whether "consumer-grade," "gaming," "server" or "workstation" or any other marketing term is applied to describe the intended end use or performance level):

In other words (and ignoring the potential unavailability of required financial resources and differences in interface technologies), if you can build a 1P machine, you can build a 2P, 4P, or 8P machine.

In this case, the 2P Opteron can be built for roughly the same price as the 1P Core i7-2600K, so as long as the determinant is capital investment dollars, the Opteron is a wiser choice *even* when building a machine intended for mainstream or gaming consumers. When server-level reliability is required, the Opteron easily wins: that's its "home-court advantage."

There isn't a similar case to be made for the Xeon: it's not junk, but the Opteron dominates the market in terms of performance per dollar and (IMHO) the Xeon offers no appreciable advantage for the sort of work you've described.

Best wishes for your build!


 


Advantages of Opterons
- Significantly less expensive than anywhere near similar Xeons, especially the quad-CPU-capable Opteron 6200s vs. the quad-CPU-capable Xeon E5-4xxx or E7-4xxx CPUs.
- Opterons are sold by core count and clock speed, everything else (feature set, cache size, RAM speed, bus speed) is identical between processors in the same model line. There are no "crippled" Opterons. Not so with Xeons, only the really expensive ones are "fully functional" with full-speed buses, full-speed memory controllers, all of the whiz-bang features enabled, the full amount of cache, etc.
- Performs much better than a similarly priced Xeon CPU especially using Linux or any other non-Windows OS, and especially on multithreaded code.
- AMD tends to support a CPU socket for more than 1-2 generations, so if you buy an AMD server board, you are more likely to be able to upgrade the CPU for a longer period of time without shelling out for a new board.
- Boards tend to be a little less expensive than equivalent Intel boards
- Greater supported memory bandwidth- all Opterons support up to DDR3-1866, while no Xeon can support more than DDR3-1600, and many multi-CPU ones only support DDR3-1333.
- Lower power consumption with quad-socket CPUs, especially the Opteron 6200s vs. Xeon E7s due to the latter's FB-DIMM2 on-motherboard memory buffers.

Advantages of Xeons
- Somewhat greater selection of motherboards available
- Better single-threaded performance than Opterons, especially with legacy Windows applications
- Can scale up to 8 CPU systems (E7-8xxx) whereas current Opterons only support up to 4 CPU systems.
- 8 CPU Xeon E7s can support more RAM per board than 4 CPU Opteron setups.
- Nobody ever got in trouble for buying Intel. Buying from the 800 lb gorilla of the market allows the IT peon to shift the blame from themselves if anything goes wrong. Buying from somebody other than the 800 lb gorilla in the field invites personal criticism and blame if anything goes wrong even if saves the CFO a bundle.
 

wireflight

Honorable
Mar 15, 2012
5
0
10,510
... but not all Opteron motherboards officially support DDR3-1866. I don't recall which, but at least either Tyan or Asus imposes a speed limit of DDR3-1066 for Socket C32 Opterons using 16-GB DIMMs (although 8-GB DIMMs get the nod for DDR3-1333, with a footnote that indicates only the 4200-series and 6200-series Opterons "are capable" of using DDR3-1600; not sure whether the same is true for Supermicro/A+.



Have you priced RAM lately? Supermicro/A+ has at least 4 different 1-TB RAM quad-socket boards for the 6000-series Opterons (SWTX form factor); half of those are made to support a 1U platform! Says Supermicro/A+ of its H8QG boards, "Supports up to 1TB Registered ECC or 256GB Unbuffered ECC/non-ECC DDR3 1600/1333/1066 SDRAM in 32 DIMMs."

Supermicro offers 2 different non-proprietary 1-TB RAM boards for the LGA2011 Xeons (and 5 proprietary 1-TB RAM boards for the LGA1567 Xeons)! Tyan only offers UP ("1S") and DP ("2S") Xeon boards. I don't know of anyone else building mainstream motherboards that can pack that much RAM, unless maybe you consider something like the MiTAC ARM Server to be mainstream: I think it goes to 1024 GB of RAM -- but technically it's 8 boards packed into a 4U solution (AMD laughs).

Perhaps your comment was intended merely to indicate the possibility that someone could theoretically build an 8P Xeon E7 motherboard and load it with more than 1TB of RAM; if so, then I agree. On the other hand, such vaporware is valuable only in shadow-boxing contests: prototypical computers rely on ideas that have been reduced to tangible reality -- so, no matter what one might imagine is possible for a maxxed-out Intel sever: in the real world, today's inexpensive Opterons easily match the greatest single-board RAM-accessing capacity of the most powerful Xeons Intel can produce.



A sobering tragedy, to be sure -- and a reminder that the bandwagon is a euphemism for mob logic. Of course, that's why deluded factions flock to the polls to support one or another candidate who has bewitched them (usually, first through party affiliation and then through shameless pandering and solicitation of favor based on appeals to prurient lusts) -- and failing those things to win support, parties and candidates work to demonize their pretend competition, perchance they might inspire some to vote against the feigned opposition. In reality, it matters not: the Democrats are neither any better nor any worse than the Republicans; in fact, they are fungible with each other, being exactly alike; each is an incorrigible enemy of both civility and humanity. Only those who are evil or insane support either in its quest for domination.

We, however, have in most cases a chance to not buy from either Intel or Microsoft. AMD has fought valiantly and honorably in delivering to us an option better than the brainless choice; similarly have labored the creators of OpenOffice and Linux and a host of other producer/providers of IT.


 


The fastest RAM that any B2-stepping Bulldozer CPU (Opteron 3200/4200/6200, AMD FX, FM2 APUs) supports is DDR3-1866, and this is ONLY supported in boards with one DIMM slot per channel and with single- or dual-ranked memory modules. There is only one Opteron board as far as I can tell which has only one DIMM slot per channel and that is Supermicro's H8QGL. I have this board and DDR3-1866 is *unofficially* supported with single-ranked or dual-ranked unbuffered or registered memory. All of the rest have two or three DIMM slots per channel and at best can support DDR3-1600 just because of the added electrical demands of having the extra traces coming from the CPU to the DIMM slots, even if you only have one DIMM per channel actually installed. This is also true for AMD FX and FM2 APU boards- four-DIMM boards top out at DDR3-1600 while two-DIMM boards can use DDR3-1866.

Using quad-ranked or low-voltage DIMMs reduces the memory speed as well and this also varies depending on the total number of DIMM slots available. There are occasionally differences between speeds supported with unbuffered DIMMs vs. registered DIMMs as well; registered DIMMs tend to support slightly higher speeds on 2 or 3 DIMM per channel boards than unbuffered. Best thing to do is consult the documentation for your board, except if it is a TYAN, in which case just go buy 1600 and see how much the board clocks it down, if any, in your specific configuration. TYAN's manuals are woefully out of date.

In case you are keeping score, Intel's servers do the same thing. The E5-4600 Sandy Bridge-EX parts support up to DDR3-1600 with one DIMM slot per channel and up to DDR3-1333 with two DIMM slots per channel using unbuffered memory even if only one slot is populated. Registered memory is DDR3-1600 with SR or DR parts either one or two DIMM slots and one or two DIMM slots populated. Of course this is the maximum speed possible, certain SKUs of Xeons cap their IMC speeds at lower than that for marketing purposes.

Have you priced RAM lately? Supermicro/A+ has at least 4 different 1-TB RAM quad-socket boards for the 6000-series Opterons (SWTX form factor); half of those are made to support a 1U platform! Says Supermicro/A+ of its H8QG boards, "Supports up to 1TB Registered ECC or 256GB Unbuffered ECC/non-ECC DDR3 1600/1333/1066 SDRAM in 32 DIMMs."

Yes, I have, it's pricey once you get beyond 8 GB modules. Even 8 GB modules are pricey if you want faster than DDR3-1333. Maximum DIMM size with current technology is 8 GB/DIMM for unbuffered RAM (two ranks of 4 Gbit 8-bit-wide ICs) and 32 GB for registered RAM (four ranks of 4 Gbit four-bit-wide ICs). 32*32 GB = 1 TB. Eventually the H8QGi/H8QG6 boards will support up to 2 TB when 8 Gbit memory ICs come out, which are the largest ICs in the DDR3 spec.

Supermicro offers 2 different non-proprietary 1-TB RAM boards for the LGA2011 Xeons (and 5 proprietary 1-TB RAM boards for the LGA1567 Xeons)! Tyan only offers UP ("1S") and DP ("2S") Xeon boards. I don't know of anyone else building mainstream motherboards that can pack that much RAM, unless maybe you consider something like the MiTAC ARM Server to be mainstream: I think it goes to 1024 GB of RAM -- but technically it's 8 boards packed into a 4U solution (AMD laughs).

Perhaps your comment was intended merely to indicate the possibility that someone could theoretically build an 8P Xeon E7 motherboard and load it with more than 1TB of RAM; if so, then I agree. On the other hand, such vaporware is valuable only in shadow-boxing contests: prototypical computers rely on ideas that have been reduced to tangible reality -- so, no matter what one might imagine is possible for a maxxed-out Intel sever: in the real world, today's inexpensive Opterons easily match the greatest single-board RAM-accessing capacity of the most powerful Xeons Intel can produce.

My comment was indeed just to indicate that theoretically one could build an 8P E7 motherboard and load it up with 2 TB of RAM. Supermicro has an 8-socket E7 setup already, the X8OBN. It takes two 4P boards and a QPI bridge card to create one single-image machine, like how TYAN used to do 8P Opteron setups.